Download 8. Cyclotomic polynomials - Math-UMN

8. Cyclotomic polynomials
8.1
8.2
8.3
8.4
8.5
8.6
Multiple factors in polynomials
Cyclotomic polynomials
Examples
Finite subgroups of fields
Infinitude of primes p = 1 mod n
Worked examples
1. Multiple factors in polynomials
There is a simple device to detect repeated occurrence of a factor in a polynomial with coefficients in a field.
Let k be a field. For a polynomial
f (x) = cn xn + . . . + c1 x + c0
with coefficients ci in k, define the (algebraic) derivative
[1]
Df (x) of f (x) by
Df (x) = ncn xn−1 + (n − 1)cn−1 xn−2 + . . . + 3c3 x2 + 2c2 x + c1
Better said, D is by definition a k-linear map
D : k[x] −→ k[x]
defined on the k-basis {xn } by
D(xn ) = nxn−1
[1.0.1] Lemma: For f, g in k[x],
D(f g) = Df · g + f · Dg
[1] Just as in the calculus of polynomials and rational functions one is able to evaluate all limits algebraically, one
can readily prove (without reference to any limit-taking processes) that the notion of derivative given by this formula
has the usual properties.
115
116
Cyclotomic polynomials
[1.0.2] Remark: Any k-linear map T of a k-algebra R to itself, with the property that
T (rs) = T (r) · s + r · T (s)
is a k-linear derivation on R.
Proof: Granting the k-linearity of T , to prove the derivation property of D is suffices to consider basis
elements xm , xn of k[x]. On one hand,
D(xm · xn ) = Dxm+n = (m + n)xm+n−1
On the other hand,
Df · g + f · Dg = mxm−1 · xn + xm · nxn−1 = (m + n)xm+n−1
yielding the product rule for monomials.
///
A field k is perfect if either the characteristic of k is 0
a1/p in k for every a ∈ k. [3]
[2]
or if, in characteristic p > 0, there is a pth root
[1.0.3] Proposition: Let f (x) ∈ k[x] with a field k, and P an irreducible polynomial in k[x]. If P e
divides f then P divides gcd(f, Df ). If k is perfect and e − 1 6= 0 in k, there is a converse:
divides both f and Df then P e divides f .
[4]
if P e−1
Proof: On one hand, suppose f = P e · g with ≥ 2. By the product rule,
Df = eP e−1 DP · g + P e · Dg
is a multiple of P e−1 .
[5]
This was the easy half.
On the other hand, for the harder half of the assertion, suppose P e−1 divides both f and Df . Write
f /P e−1 = Q · P + R
with deg R < deg P . Then f = QP e + RP e−1 . Differentiating,
Df = DQ P e + eQP e−1 DP + DR P e−1 + R(e − 1)P e−2 DP
By hypothesis P e−1 divides Df . All terms on the right-hand side except possibly R(e − 1)P e−2 DP are
divisible by P e−1 , so P divides R(e − 1)P e−2 DP . Since P is irreducible, either e − 1 = 0 in k, or P divides
R, or P divides DP . If P divides R, P e divides f , and we’re done.
If P does not divide R then P divides DP . Since deg DP < deg P , if P divides DP then DP = 0. This
would require that all the exponents of x occurring with non-zero coefficient are divisible by the characteristic
p, which must be positive. So P is of the form
P (x) = apm xpm + ap(m−1) xp(m−1) + ap(m−2) xp(m−2) + . . . + a2p x2p + ap xp + a0
[2] as for
Q, R, and C
[3] As is the case for finite fields such as
Z/p, by Fermat’s Little Theorem.
[4] In particular, this converse holds if the characteristic of k is 0.
[5] This half does not need the irreducibility of P .
Garrett: Abstract Algebra
117
Using the perfect-ness of the field k, each ai has a pth root bi in k. Because the characteristic is p > 0,
(A + B)p = Ap + B p
Thus, P (x) is the pth power of
bpm xn + bp(m−1) x(m−1) + bp(m−2) x(m−2) + . . . + b2p x2 + bp x + b0
If P is a pth power it is not irreducible. Therefore, for P irreducible DP is not the zero polynomial. Therefore,
R = 0, which is to say that P e divides f , as claimed.
///
2. Cyclotomic polynomials
For b 6= 0 in a field k, the exponent of b is the smallest positive integer n (if it exists) such that bn = 1.
That is, b is a root of xn − 1 but not of xd − 1 for any smaller d. We construct polynomials Φn (x) ∈ Z[x]
such that
Φn (b) = 0 if and only if b is of exponent n
These polynomials Φn are cyclotomic polynomials.
[2.0.1] Corollary: The polynomial xn − 1 has no repeated factors in k[x] if the field k has characteristic
not dividing n.
Proof: It suffices to check that xn − 1 and its derivative nxn−1 have no common factor. Since the
characteristic of the field does not to divide n, n · 1k 6= 0 in k, so has a multiplicative inverse t in k,
and
(xn − 1) − (tx) · (nxn−1 ) = −1
and gcd(xn − 1, nxn−1 ) = 1.
///
Define the nth cyclotomic polynomial Φn (x) by
Φ1 (x) = x − 1
and for n > 1, inductively,
Φn (x) =
xn − 1
lcm of all xd − 1 with 0 < d < n, d dividing n
with the least common multiple monic.
[2.0.2] Theorem:
• Φn is a monic polynomial with integer coefficients. [6]
• For α in the field k, Φn (α) = 0 if and only if αn = 1 and αt 6= 1 for all 0 < t < n.
• gcd(Φm (x), Φn (x)) = 1 for m < n with neither m nor n divisible by the characteristic of the field k.
• The degree of Φn (x) is ϕ(n) (Euler’s phi-function)
• Another description of Φn (x):
xn − 1
Φn (x) = Q
1≤d<n,d|n Φd (x)
[6] More properly, if the ambient field k is of characteristic 0, then the coefficients lie in the copy of
Z
inside the
prime field
inside k. If the ambient field is of positive characteristic, then the coefficients lie inside the prime field
(which is the natural image of in k). It would have been more elegant to consider the cyclotomic polynomials as
polynomials in [x], but this would have required that we wait longer.
Q
Z
Z
118
Cyclotomic polynomials
• xn − 1 factors as
xn − 1 =
Y
Φd (x)
1≤d≤n,d|n
Proof: We know that d|n (and d > 0) implies that xd − 1 divides xn − 1. Therefore, by unique factorization,
the least common multiple of a collection of things each dividing xn − 1 also divides xn − 1. Thus, the
indicated lcm does divide xn − 1.
For α in k, x − α divides Φn (x) if and only if Φn (α) = 0. And αt = 1 if and only if x − α divides xt − 1.
The definition
xn − 1
Φn (x) =
lcm of all xd − 1 with 0 < d < n, d dividing n
shows first that Φn (α) = 0 implies αn = 1. Second, if αt = 1 for any proper divisor t of n then x − α divides
xt − 1, and thus x − α divides the denominator. But xn − 1 has no repeated factors, so x − α dividing the
denominator would prevent x − α dividing Φn (x), contradiction. That is, Φn (α) = 0 if and only if α is of
order n.
To determine the gcd of Φm and Φn for neither m nor n divisible by the characteristic of k, note that Φm
divides xm − 1 and Φn divides xn − 1, so
gcd(Φm , Φn ) divides gcd(xm − 1, xn − 1)
We claim that for m, n two integers (divisible by the characteristic or not)
gcd(xm − 1, xn − 1) = xgcd(m,n) − 1
Prove this claim by induction on the maximum of m and n. Reduce to the case m > n, wherein
xm − 1 − xm−n · (xn − 1) = xm−n − 1
For g a polynomial dividing both xm − 1 and xn − 1, g divides xm−n − 1. By induction,
gcd(xm−n − 1, xn − 1) = xgcd(m−n,n) − 1
But
gcd(m, n) = gcd(m − n, n)
and
xm − 1 = xm−n · (xn − 1) + xm−n − 1
so
gcd(xm − 1, xn − 1) = gcd(xm−n − 1, xn − 1)
and induction works. Thus,
gcd(xm − 1, xn − 1) = xgcd(m,n) − 1
Since
d≤m<n
d is a proper divisor of n. Thus, from
Φn (x) =
xn − 1
lcm of all x − 1 with 0 < d < n, d dividing n
d
we see that Φn (x) divides (xn − 1)/(xd − 1). Since xn − 1 has no repeated factors, Φn (x) has no factors in
common with xd − 1. Thus, gcd(Φm , Φn ) = 1.
Garrett: Abstract Algebra
119
Next, use induction to prove that
Y
xn − 1 =
Φd (x)
1≤d≤n, d|n
For n = 1 the assertion is true. From the definition of Φn ,
xn − 1 = Φn (x) · lcm{xd − 1 : d|n, 0 < d < n}
By induction, for d < n
Y
xd − 1 =
Φe (x)
0<e≤d,e|d
Since for m < n the gcd of Φm and Φn is 1,
Y
lcm{xd − 1 : d|n, 0 < d < n} =
Φd (x)
d|n,d<n
Thus,
xn − 1 = Φn (x) ·
Y
Φd (x)
d|n,d<n
as claimed.
Inductively, since all lower-index cyclotomic polynomials have integer coefficients [7] and are monic, and
xn − 1 is monic with integer coefficients, the quotient of xn − 1 by the product of the lower ones is monic
with integer coefficients.
The assertion about the degree of Φn follows from the identity (see below) for Euler’s phi-function
X
ϕ(d) = n
d|n,d>0
This completes the proof of the theorem.
///
[2.0.3] Proposition: Let ϕ(x) be Euler’s phi-function
X
ϕ(x) =
1
1≤`≤x;gcd(`,x)=1
Then for m and n relatively prime
ϕ(mn) = ϕ(m) · ϕ(n)
(weak multiplicativity)
For p prime and ` a positive integer
ϕ(p` ) = (p − 1) · p`−1
And
X
ϕ(d) = n
d|n,d>0
Proof: By unique factorization, for gcd(m, n) = 1,
gcd(t, mn) = gcd(t, m) · gcd(t, n)
[7] Or, more properly, coefficients in the canonical image of
Z in the field k.
120
Cyclotomic polynomials
so, t is prime to mn if and only if t is prime to both m and n. The gcd of m and n is the smallest positive
integer of the form rm + sn. By Sun-Ze,
f : Z/m ⊕ Z/n −→ Z/mn
by
f : (x, y) −→ rmy + snx
is a bijection, since m and n are coprime. From rm + yn = 1, rm = 1 mod n so rm is prime to n, and
sn = 1 mod m so sn is prime to m. Thus, rmy + snx has a common factor with m if and only if x does,
and rmy + snx has a common factor with n if and only if y does. Thus, f gives a bijection
{x : 1 ≤ x < m, gcd(x, m) = 1)} × {y : 1 ≤ y < n, gcd(y, n) = 1)}
−→ {z : 1 ≤ z < mn, gcd(z, mn) = 1)}
and ϕ(mn) = ϕ(m) · ϕ(n). This reduces calculation of ϕ() to calculation for prime powers pe . An integer x
in 1 ≤ x < pe is prime to pe if and only if it is not divisible by p, so there are
ϕ(pe ) = pe − pe−1 = (p − 1)pe−1
such x, as claimed.
To prove
X
ϕ(d) = n
d|n,d>0
start with n a prime power pe , in which case
X
ϕ(d) =
d|pe
X
X
ϕ(pk ) = 1 +
0≤k≤e
(p − 1)pk−1 = 1 + (p − 1)(pe − 1)/(p − 1) = pe
1≤k≤e
Let n = pe11 . . . pet t with distinct primes pi . Then


X
ϕ(d) =
Y
X

e
i=1,...,t
d|n
d|pi i
ϕ(d) =
Y
Y
ϕ(pei i ) = ϕ(
pei i ) = ϕ(n)
i=1,...,t
i
This proves the desired identity for ϕ.
///
3. Examples
For prime p, the factorization of xp − 1 into cyclotomic polynomials is boring
Φp (x) =
xp − 1
= xp−1 + xp−2 + . . . + x2 + x + 1
x−1
For n = 2p with odd prime p
Φ2p (x) =
x2p − 1
x2p − 1
xp + 1
=
=
Φ1 (x) Φ2 (x) Φp (x)
Φ2 (x) (xp − 1)
x+1
= xp−1 − xp−2 + xp−3 − . . . + x2 − x + 1
Garrett: Abstract Algebra
121
For n = p2 with p prime,
2
2
xp − 1
xp − 1
Φp2 (x) =
= p
= xp(p−1) + xp(p−2) + . . . + xp + 1
Φ1 (x) Φp (x)
x −1
Generally, one observes that for n = pe a prime power
Φpe (x) = Φp (xp
e−1
) = xp
e−1
(p−1)
+ xp
e−1
(p−2)
+ . . . + x2p
e−1
+ x2p
e−1
+1
For n = 2 · m (with odd m > 1) we claim that
\P hi2m (x) = Φm (−x)
Note that, anomalously, Φ2 (−x) = −x + 1 = −Φ1 (x). Prove this by induction:
Φ2m (x) = Q
d|m
x2m − 1
x2m − 1
Q
Q
= m
(x − 1) d|m, d<m Φd (−x)
Φd (x) · d|m, d<m Φ2d (x)
xm + 1
(xm + 1) Φm (−x)
=
= Φm (−x)
((−x)m − 1) · (−1)
d|m, d<m Φd (−x)
=Q
by induction, where the extra −1 in the denominator was for Φ2 (−x) = −Φ1 (x), and (−1)m = −1 because
m is odd.
Thus,
Φ3 (x) =
x3 − 1
= x2 + x + 1
x−1
Φ9 (x) = Φ3 (x3 ) = x6 + x3 + 1
Φ18 (x) = Φ9 (−x) = x6 − x3 + 1
For n = pq with distinct primes p, q some unfamiliar examples appear.
Φ15 (x) =
x15 − 1
x15 − 1
x15 − 1
=
=
Φ1 (x)Φ3 (x)Φ5 (x)
Φ1 (x)Φ3 (x)Φ5 (x)
Φ3 (x)(x5 − 1)
=
by direct division
[8]
x10 + x5 + 1
= x8 − x7 + x5 − x4 + x3 − x + 1
x2 + x + 1
at the last step. And then
Φ30 (x) = Φ15 (−x) = x8 + x7 − x5 − x4 − x3 + x + 1
[3.0.1] Remark: Based on a few hand calculations, one might speculate that all coefficients of all
cyclotomic polynomials are either +1, −1, or 0, but this is not true. It is true for n prime, and for n having
at most 2 distinct prime factors, but not generally. The smallest n where Φn (x) has an exotic coefficient
seems to be n = 105 = 3 · 5 · 7.
Φ105 (x) =
x105 − 1
Φ1 (x)Φ3 (x)Φ5 (x)Φ7 (x)Φ15 (x)Φ21 (x)Φ35 (x)
[8] Only mildly painful. Any lesson to be learned here?
122
Cyclotomic polynomials
=
x70 + x35 + 1
(x70 + x35 + 1)(x7 − 1)
x105 − 1
=
=
35
Φ3 (x)Φ15 (x)Φ21 (x)(x − 1)
Φ3 (x)Φ15 (x)Φ21 (x)
Φ15 (x)(x21 − 1)
=
(x70 + x35 + 1)(x7 − 1)Φ1 (x)Φ3 (x)Φ5 (x)
(x15 − 1)(x21 − 1)
=
(x70 + x35 + 1)(x7 − 1)(x5 − 1)Φ3 (x)
(x15 − 1)(x21 − 1)
Instead of direct polynomial computations, we do power series [9] computations, imagining that |x| < 1, for
example. Thus,
−1
1
= 1 + x21 + x42 + x63 + . . .
=
x21 − 1
1 − x21
We anticipate that the degree of Φ105 (x) is (3 − 1)(5 − 1)(7 − 1) = 48. We also observe that the coefficients
of all cyclotomic polynomials are the same back-to-front as front-to-back (why?). Thus, we’ll use power
series in x and ignore terms of degree above 24. Thus,
Φ105 (x) =
(x70 + x35 + 1)(x7 − 1)(x5 − 1)(x2 + x + 1)
(x15 − 1)(x21 − 1)
= (1 + x + x2 )(1 − x7 )(1 − x5 )(1 + x15 )(1 + x21 )
= (1 + x + x2 ) × (1 − x5 − x7 + x12 + x15 − x20 + x21 − x22 )
= 1 + x + x2 − x5 − x6 − x7 − x7 − x8 − x9 + x12 + x13 + x14 + x15 + x16 + x17
−x20 − x21 − x22 + x21 + x22 + x23 − x22 − x23 − x24
= 1 + x + x2 − x5 − x6 − 2x7 − x8 − x9 + x12 + x13 + x14
+x15 + x16 + x17 − x20 − x22 − x24
Looking closely, we have a −2x7 .
Less well known are Lucas-Aurifeullian-LeLasseur factorizations such as
x4 + 4 = (x4 + 4x2 + 4) − (2x)2 = (x2 + 2x + 2)(x2 − 2x + 2)
More exotic are
x6 + 27
= (x2 + 3x + 3)(x2 − 3x + 3)
x2 + 3
x10 − 55
= (x4 + 5x3 + 15x2 + 25x + 25) × (x4 − 5x3 + 15x2 − 25x + 25)
x2 − 5
and
and further
x12 + 66
= (x4 + 6x3 + 18x + 36x + 36) × (x4 − 6x3 + 18x − 36x + 36)
x4 + 36
x14 + 77
= (x6 + 7x5 + 21x4 + 49x3 + 147x2 + 343x + 343)
x2 + 7
× (x6 − 7x5 + 21x4 − 49x3 + 147x2 − 343x + 343)
[9] In fact, one is not obliged to worry about convergence, since one can do computations in a formal power series ring.
Just as polynomials can be precisely defined by their finite sequences of coefficients, with the obvious addition and
multiplication mirroring our intent, formal power series are not-necessarily-finite sequences with the same addition
and multiplication, noting that the multiplication does not require any infinite sums. The formal adjective here
merely indicates that convergence is irrelevant.
Garrett: Abstract Algebra
123
The possibility and nature of these factorizations are best explained by Galois theory.
4. Finite subgroups of fields
Now we can prove that the multiplicative group k × of a finite field k is a cyclic group. When k is finite, a
generator of k × is a primitive root for k.
[4.0.1] Theorem: Let G be a finite subgroup of k× for a field k. Then G is cyclic.
[4.0.2] Corollary: For a finite field k, the multiplicative group k× is cyclic.
///
[10]
any element of G is a root of the polynomial f (x) = xn − 1.
We know that a polynomial with coefficients in a field k has at most as many roots (in k) as its degree, so
this polynomial has at most n roots in k. Therefore, it has exactly n roots in k, namely the elements of the
subgroup G.
Proof: Let n be the order of G. Then
The characteristic p of k cannot divide n, since if it did then the derivative of f (x) = xn − 1 would be zero,
and gcd(f, f 0 ) = f and f would have multiple roots. Thus,
Y
xn − 1 =
Φd (x)
d|n
Since xn − 1 has n roots in k, and since the Φd ’s here are relatively prime to each other, each Φd with d|n
must have a number of roots (in k) equal to its degree. Thus, Φd for d|q − 1 has ϕ(d) > 0 roots in k (Euler’s
phi-function).
The roots of Φn (x) are b ∈ k × such that bn = 1 and no smaller positive power than n has this property.
Any root of Φn (x) = 0 in k × would be a generator of the (therefore cyclic) group G. The cyclotomic
polynomial Φn has ϕ(n) > 0 zeros, so G has a generator, and is cyclic.
///
5. Infinitude of primes p = 1 mod n
This is a very special case of Dirichlet’s theorem that, given a modulus n and a fixed integer a relatively
prime to n, there are infinitely-many primes p = a mod n. We only treat the case a = 1.
[5.0.1] Corollary: Fix 1 < n ∈ Z. There are infinitely many primes p = 1 mod n.
Proof: Recall that the nth cyclotomic polynomial Φn (x) is monic (by definition), has integer coefficients,
and has constant coefficient ±1. [11] And Φn (x) is not constant. Suppose there were only finitely-many
primes p1 , . . . , pt equal to 1 mod n. Then for large-enough positive integer `,
N = Φn (` · np1 . . . pt ) > 1
and N is an integer. Since Φn (x) has integer coefficients and has constant term ±1, for each pi we have
N = ±1 mod pi , so in particular no pi divides N . But since N > 1 it does have some prime factor p. Further,
since the constant term is ±1, N = ±1 mod n, so p is relatively prime to n. Then
Φn (` · np1 . . . pt ) = N = 0 mod p
[10] Lagrange, again.
[11] The assertion about the constant coefficient follows from the fact that Φ (x) is monic, together with the fact that
n
Φ(x−1 ) = ±Φ(x), which is readily proven by induction.
124
Cyclotomic polynomials
×
Thus, ` · np1 . . . pt has order n in F×
p . By Lagrange, n divides |Fp | = p − 1, so p = 1 mod n. Contradiction
to the finiteness assumption, [12] so there are infinitely-many primes p = 1 mod n.
///
6. Worked examples
[8.1] Gracefully verify that the octic x8 + x7 + x6 + x5 + x4 + x3 + x2 + x + 1 factors properly in Q[x].
This octic is
x9 − 1
x3 − 1)(x6 + x3 + 1)
=
= (x2 + x + 1) (x6 + x3 + 1)
x−1
x−1
for example. We might anticipate this reducibility by realizing that
x9 − 1 = Φ1 (x) Φ3 (x) Φ9 (x)
where Φn is the nth cyclotomic polynomial, and the given octic is just (x9 − 1)/Φ1 (x), so what is left at least
factors as Φ3 (x) Φ9 (x).
[8.2] Gracefully verify that the quartic x4 + x3 + x2 + x + 1 is irreducible in F2 [x].
Use the recursive definition of cyclotomic polynomials
Φn (x) = Q
xn − 1
d|n, d<n Φd (x)
Thus, the given quartic is Φ5 (x). And use the fact that for the characteristic of the field k not dividing n,
Φn (α) = 0 if and only if α is of order n in k × . If it had a linear factor x − α with α ∈ F2 , then Φ4 (α) = 0,
×
and α would be of order 5 in F×
2 . But F2 is of order 1, so has no elements of order 5 (by Lagrange). (We
saw earlier that) existence of an irreducible quadratic factor of Φ4 (x) in F2 [x] is equivalent to existence of
×
2
an element α of order 5 in F×
22 , but |F22 | = 2 − 1 = 3, which is not divisible by 5, so (Lagrange) has no
element of order 5. The same sort of argument would show that there is no irreducible cubic factor, but we
already know this since if there were any proper factorization then there would be a proper factor of at most
half the degree of the quartic. But there is no linear or quadratic factor, so the quartic is irreducible.
[8.3] Gracefully verify that the sextic x6 + x5 + x4 + x3 + x2 + x + 1 is irreducible in F3 [x].
Use the recursive definition of cyclotomic polynomials
Φn (x) = Q
xn − 1
d|n, d<n Φd (x)
Thus, the given sextic is Φ7 (x). And use the fact that for the characteristic of the field k not dividing n,
Φn (α) = 0 if and only if α is of order n in k × . If it had a linear factor x − α with α ∈ F3 , then Φ7 (α) = 0,
×
and α would be of order 7 in F×
2 . But F3 is of order 2, so has no elements of order 7 (Lagrange). Existence
of an (irreducible) quadratic factor of Φ7 (x) in F3 [x] is equivalent to existence of an element α of order 7 in
F×32 , but |F×32 | = 32 − 1 = 8, which is not divisible by 7, so (Lagrange) has no element of order 5. Similarly,
if there were an (irreducible) cubic factor, then there would be a root in a cubic extension F33 of F3 , but
F×33 has order 33 − 1 = 26 which is not divisible by 7, so there is no such element. If there were any proper
[12] Mildly ironic that we have a contradiction, considering that we seem to have just succeeded in proving that there
is one more prime of the type that we want. Perhaps this suggests that it is needlessly inefficient to couch this
argument as proof by contradiction.
Garrett: Abstract Algebra
125
factorization then there would be a proper factor of at most half the degree of the sextic. But there is no
linear, quadratic, or cubic factor, so the sextic is irreducible.
[8.4] Gracefully verify that the quartic x4 + x3 + x2 + x + 1 in factors into two irreducible quadratics in
F19 [x].
As above, we see that the quartic is the 5th cyclotomic polynomial. If it had a linear factor in F19 [x] then
(since the characteristic 19 does not divide the index 5) there would be an element of order 5 in F×
19 , but the
latter group has order 19 − 1 not divisible by 5, so (Lagrange) there is no such element. But the quadratic
extension F192 of F19 has multiplicative group with order 192 − 1 = 360 which is divisible by 5, so there is
an element α of order 5 there.
Since α ∈ F192 − F19 , the minimal polynomial M (x) of α over F19 is quadratic. We have shown that in this
circumstance the polynomial M divides the quartic. (Again, the proof is as follows: Let
x4 + x3 + x2 + x + 1 = Q(x) · M (x) + R(x)
with Q, R ∈ F19 [x] and deg R < deg M . Evaluating at α gives R(α) = 0, which (by minimality of M ) implies
R is the 0 polynomial. Thus, M divides the quartic.) The quotient of the quartic by M is quadratic, and
(as we’ve already seen) has no linear factor in F19 [x], so is irreducible.
[8.5] Let f (x) = x6 − x3 + 1. Find primes p with each of the following behaviors: f is irreducible in Fp [x], f
factors into irreducible quadratic factors in
into linear factors in Fp [x].
Fp [x], f factors into irreducible cubic factors in Fp [x], f factors
By the recursive definition and properties of cyclotomic polynomials, we recognize f (x) as the 18th cyclotomic
polynomial Φ18 (x). For a prime p not dividing 18, zeros of Φ18 are exactly elements of order 18. Thus, if
pd − 1 = 0 mod 18 but no smaller exponent than d achieves this effect, then F×
(proven cyclic by now) has
pd
an element of order 18, whose minimal polynomial divides Φ18 (x).
We might observe that (Z/18)× is itself cyclic, of order ϕ(18) = ϕ(2)ϕ(32 ) = (3 − 1)3 = 6, so has elements
of all possible orders, namely 1, 2, 3, 6.
For p = 1 mod 18, for example p = 19, already p − 1 = 0 mod 18, so f (x) has a linear factor in
is the case of order 1 element in (Z/18)× .
F19 [x]. This
A moment’s thought might allow a person to realize that 17 = −1 is an element (and the only element) of
order 2 in (Z/18)× . So any prime p = 17 mod 18 (for example p = 17 itself, by coincidence prime) will have
the property that F×
p2 has elements of order 18. Indeed, by properties of cyclic groups, it will have ϕ(18) = 6
elements of order 18 there, each of whose minimal polynomial is quadratic. Thus (since a quadratic has at
most two zeros) there are at least 3 irreducible quadratics dividing the sextic Φ18 (x) in Fp [x]. Thus, since
degrees add in products, these three quadratics are all the factors of the sextic.
After a bit of trial and error, one will find an element of order 3 in (Z/18)× , such as 7. Thus, for p = 7 mod 18
(such as 7 itself, which by coincidence is prime), there is no element of order 18 in Fp or in Fp2 , but there is
one in Fp3 , whose minimal polynomial over Fp is therefore cubic and divides Φ18 . Again, by properties of
cyclic groups, there are exactly ϕ(18) = 6 such elements in Fp3 , with cubic minimal polynomials, so there
are at least (and, thus, exactly) two different irreducible cubics in Fp [x] dividing Φ18 (x) for such p.
After a bit more trial and error, one finds an element of order 6 in (Z/18)× , such as 5. (The other is 11.)
Thus, for p = 5 mod 18 (such as 5 itself, which by coincidence is prime), there is no element of order 18 in
Fp or in Fp2 , or Fp3 , but there is one in Fp6 . (By Lagrange, the only possible orders of p in (Z/18)× are
1, 2, 3, 6, so we need not worry about p4 or p5 ). The minimal polynomial of such an element is Φ18 (x), which
is (thus, necessarily) irreducible in Fp [x].
[8.6] Explain why x4 + 1 properly factors in Fp [x] for any prime p.
126
Cyclotomic polynomials
As in the previous problems, we observe that x4 + 1 is the 8th cyclotomic polynomial. If p|8, namely p = 2,
then this factors as (x − 1)4 . For odd p, if p = 1 mod 8 then F×
p , which we now know to be cyclic, has an
element of order 8, so x4 + 1 has a linear factor. If p 6= 1 mod 8, write p = 2m + 1, and note that
p2 − 1 = (2m + 1)2 − 1 = 4m2 + 4m = m(m + 1) · 4
so, if m is odd, m + 1 is even and p2 − 1 = 0 mod 8, and if m is even, the same conclusion holds. That is, for
odd p, p2 − 1 is invariably divisible by 8. That is, (using the cyclicness of any finite field) there is an element
of order 8 in Fp2 . The minimal polynomial of this element, which is quadratic, divides x4 + 1 (as proven in
class, with argument recalled above in another example).
[8.7] Explain why x8 − x7 + x5 − x4 + x3 − x + 1 properly factors in Fp [x] for any prime p. (Hint: It factors
either into linear factors, irreducible quadratics, or irreducible quartics.)
The well-read person will recognize this octic as Φ15 (x), the fifteenth cyclotomic polynomial. For a prime
p not dividing 15, zeros of Φ1 5 in a field Fpd are elements of order 15, which happens if and only if
pd − 1 = 0 mod 15, since we have shown that F×
is cyclic. The smallest d such that pd = 1 mod 15 is the
pd
×
order of p in (Z/15) . After some experimentation, one may realize that (Z/15)× is not cyclic. In particular,
every element is of order 1, 2, or 4. (How to see this?) Granting this, for any p other than 3 or 5, the minimal
polynomial of an order 15 element is linear, quadratic, or quartic, and divides Φ15 .
For p = 3, there is some degeneration, namely x3 − 1 = (x − 1)3 . Thus, in the (universal) expression
Φ15 (x) =
x15 − 1
Φ1 (x) Φ3 (x) Φ5 (x)
we actually have
Φ15 (x) =
(x5 − 1)2
(x5 − 1)3
=
= (x4 + x3 + x2 + 1)2
(x − 1)2 (x5 − 1)
(x − 1)2
For p = 5, similarly, x5 − 1 = (x − 1)5 , and
Φ15 (x) =
x15 − 1
(x3 − 1)5
(x3 − 1)4
= 3
=
= (x2 + x + 1)4
Φ1 (x) Φ3 (x) Φ5 (x)
(x − 1) (x − 1)4
(x − 1)4
[8.8] Why is x4 − 2 irreducible in F5 [x]?
A zero of this polynomial would be a fourth root of 2. In F×
5 , one verifies by brute force that 2 is of order
4, so is a generator for that (cyclic) group, so is not a square in F×
5 , much less a fourth power. Thus, there
is no linear factor of x4 − 2 in F5 [x].
4
4
16
2
The group F×
= 1.
52 is cyclic of order 24. If 2 were a fourth power in F5 , then 2 = α , and 2 = 1 gives α
24
8
Also, α = 1 (Lagrange). Claim that α = 1: let r, s ∈ Z be such that r · 16 + s · 24 = 8, since 8 is the
greatest common divisor. Then
α8 = α16r+24s = (α16 )r · (α24 )s = 1
This would imply
22 = (α4 )2 = α8 = 1
which is false. Thus, 2 is not a fourth power in
F5 , so the polynomial x4 − 2 has no quadratic factors.
2
A quartic with no linear or quadratic factors is irreducible (since any proper factorization of a polynomial
P must involve a factor of degree at most half the degree of P ). Thus, x4 − 2 is irreducible in F5 [x].
[8.9] Why is x5 − 2 irreducible in F11 [x]?
Garrett: Abstract Algebra
127
As usual, to prove irreducibility of a quintic it suffices to show that there are no linear or quadratic factors.
To show the latter it suffices to show that there is no zero in the underlying field (for linear factors) or in a
quadratic extension (for irreducible quadratic factors).
First determine the order of 2 in F11 : since |F×
11 | = 10, it is either 1, 2, 5, or 10. Since 2 6= 1 mod 11, and
22 − 1 = 3 6= 0 mod 11, and 25 − 1 = 31 6= 0 mod 11, the order is 10. Thus, in F11 it cannot be that 2 is a
fifth power.
2
5
10
The order of F×
= 1 imply
112 is 11 − 1 = 120. If there were a fifth root α of 2 there, then α = 2 and 2
50
120
α = 1. Also, (Lagrange) α
= 1. Thus, (as in the previous problem) α has order dividing the gcd of 50
and 120, namely 10. Thus, if there were such α, then
22 = (α5 )2 = α10 = 1
But 22 6= 1, so there is no such α.
Exercises
8.[6.0.1] Determine the coefficients of the 12th cyclotomic polynomial.
8.[6.0.2] Gracefully verify that (x15 − 1)/(x5 − 1) factors properly in Q[x].
8.[6.0.3] Find a prime p such that the 35th cyclotomic polynomial has an irreducible 12th -degree factor
in
Fp [x].
8.[6.0.4] Determine the factorization into irreducibles of (x7 − 1)/(x − 1) in F2 [x].
8.[6.0.5] Explain why the 12th cyclotomic polynomial factors properly in Fp [x] for any prime p.
8.[6.0.6] Explain why the thirty-fifth cyclotomic polynomial factors properly in Fp [x] for any prime p.
8.[6.0.7] Show that a finite field extension of Q contains only finitely-many roots of unity.
8.[6.0.8] Let p be a prime and n ≥ 1. Let ϕm be the mth cyclotomic polynomial. Show that
ϕpn (x) =

p

 ϕn (x )
(for p|n)
p

 ϕn (x )
ϕn (x)
(otherwise)
8.[6.0.9] Let n = 2a pb qc for primes p, q. Show that the coefficients of the cyclotomic polynomial ϕn are
in the set {−1, 0, 1}.
8.[6.0.10] Suppose n is divisible by p2 for some prime p. Show that the sum of the primitive nth roots of
unity is 0.